Stop Renting AI Image Tools: How to Run Every Model Yourself with Claude Code + Replicate

If you’ve paid for an AI image or video tool lately, you’ve felt the shape of the market: a monthly plan, a generation cap, and a nagging sense that you’re renting something you use every day. Search “best AI video generator” or “AI image tool” and you get a price list, not an answer — everyone comparing subscriptions, nobody mentioning that the models underneath are the same across most of them.

This piece walks through a different setup. You connect Claude directly to a pay-per-use model engine, and one connection gives you every major AI image and video model — driven in plain language, billed by the generation, with no subscription. I’ll keep the deep step-by-step for the full guide, but by the end of this you’ll understand exactly how it works and why it’s cheaper.

The engine room

The popular tools are a dashboard wrapped around a handful of models: Veo, Seedance and Kling for video; OpenAI’s and Google’s models for images. Those models are reachable directly. Two platforms make that easy by putting hundreds of them behind one connection: Replicate (tens of thousands of models, pay-as-you-go, now part of Cloudflare) and fal.ai (around a thousand image and video models under one API — and, per the State of Generative Media report, the engine behind roughly half of all image generation and 44% of video generation today).

Hugging Face is the cousin people know — but it’s mainly a library and community. Replicate and fal are built to actually run a model on demand and hand you the file. That’s the difference that matters here.

Drive it with Claude

You don’t call these models directly. You give Claude the key once and then you ask: “six ad creatives from this product photo,” “a 12-second vertical reel of this scene,” “on-brand thumbnails with this headline.” Claude works out which model fits, runs it, and gives you the result to review. Describe, run, refine — that’s the whole loop, and none of it is code.

The economics

An image runs about $0.07–0.20. A short video runs about $2–4. A heavy month — a hundred images and ten videos — is roughly $30–50, against €40/90/200 for a capped subscription. Generate nothing and you pay nothing. This is exactly what people keep asking for: on Reddit you’ll find thread after thread — “how are you making AI videos cheaply,” “a free image generator with no subscription” — and one developer’s blunt post, “AI image generation APIs were too expensive, so I built my own.” The demand is real; this is the answer.

You own it

A subscription can change price or limits tomorrow, and your workflow lives inside it. Here, you use the same models, plugged in directly, through a connection you control. A better model launches next month? You point Claude at it the same day. Nothing to migrate, nothing to renegotiate. And a fair word of honesty: “free” means no subscription, not zero — you pay per generation, and you still judge results with your own eye. That’s the trade, and for anyone generating regularly it’s a good one.

The full no-code walkthrough — the setup, the which-model-for-which-job map, the prompt library, and the research-to-creative chain — is here:

• 🛒 Do it yourself → https://webproitalia.gumroad.com/l/aiimage